diff --git a/frameworks/native/neural_network_core/cpp_type.h b/frameworks/native/neural_network_core/cpp_type.h
index faf90394b9ff8aaa1e0585c2478ef74567ea4418..7007244d397071c6d08488f3ea2cbecb24309c61 100644
--- a/frameworks/native/neural_network_core/cpp_type.h
+++ b/frameworks/native/neural_network_core/cpp_type.h
@@ -34,6 +34,13 @@ enum DeviceStatus: int {
     OFFLINE
 };
 
+enum class TuningStrategy {
+    OFF = 0,
+    ON_DEVICE_TUNING,
+    ON_DEVICE_PREPROCESS_TUNING,
+    ON_CLOUD_TUNING
+};
+
 struct ModelConfig {
     bool enableFloat16;
     OH_NN_PerformanceMode mode;
@@ -41,6 +48,7 @@ struct ModelConfig {
     std::string isProfiling;
     std::string cachePath;
     std::map<std::string, std::string> opLayout;
+    TuningStrategy tuningStrategy{TuningStrategy::OFF};
 };
 
 struct Buffer {
diff --git a/frameworks/native/neural_network_runtime/inner_model.cpp b/frameworks/native/neural_network_runtime/inner_model.cpp
index b9dfea9c1208489ad596d6a82fc5c30ee4fc84e7..15f51e234ea35a83db63e8e0f3f9424f6b3a2378 100644
--- a/frameworks/native/neural_network_runtime/inner_model.cpp
+++ b/frameworks/native/neural_network_runtime/inner_model.cpp
@@ -780,5 +780,15 @@ std::map<std::string, std::string> InnerModel::GetOpLayouts() const
 {
     return m_opLayouts;
 }
+
+TuningStrategy InnerModel::GetTuningStrategy() const
+{
+    return m_tuningStrategy;
+}
+
+void InnerModel::SetTuningStrategy(const TuningStrategy tuningStrategy)
+{
+    m_tuningStrategy = tuningStrategy;
+}
 }  // namespace NeuralNetworkRuntime
 }  // namespace OHOS
diff --git a/frameworks/native/neural_network_runtime/inner_model.h b/frameworks/native/neural_network_runtime/inner_model.h
index e5cc36f9c10b26c4e957e856982167c28dc0f6b7..a5c3fb068b18ce5baa95a0758b0b8c80e90c487e 100644
--- a/frameworks/native/neural_network_runtime/inner_model.h
+++ b/frameworks/native/neural_network_runtime/inner_model.h
@@ -63,6 +63,8 @@ public:
     std::string GetModelName() const;
     std::string GetProfiling() const;
     std::map<std::string, std::string> GetOpLayouts() const;
+    TuningStrategy GetTuningStrategy() const;
+    void SetTuningStrategy(const TuningStrategy tuningStrategy);
 
 private:
     void AddTensorsToLiteGraph(std::unordered_map<uint32_t, uint32_t>& modelIDToGraphID);
@@ -86,6 +88,7 @@ private:
     std::string m_modelName;
     std::string m_isProfiling;
     std::map<std::string, std::string> m_opLayouts;
+    TuningStrategy m_tuningStrategy{TuningStrategy::OFF};
 };
 }  // namespace NeuralNetworkRuntime
 }  // namespace OHOS
diff --git a/frameworks/native/neural_network_runtime/neural_network_runtime.cpp b/frameworks/native/neural_network_runtime/neural_network_runtime.cpp
index 020ab6a95a496cc3921fc1190f0196f2d47707b1..f6402f61165e2dfe016e824f14e1d7fb6d538db9 100644
--- a/frameworks/native/neural_network_runtime/neural_network_runtime.cpp
+++ b/frameworks/native/neural_network_runtime/neural_network_runtime.cpp
@@ -331,6 +331,7 @@ NNRT_API OH_NN_ReturnCode OH_NNModel_BuildFromLiteGraph(OH_NNModel *model, const
 
     auto *pLiteGraph = static_cast<const mindspore::lite::LiteGraph*>(liteGraph);
     InnerModel *innerModel = reinterpret_cast<InnerModel*>(model);
+    innerModel->SetTuningStrategy(TuningStrategy::ON_DEVICE_PREPROCESS_TUNING);
 
     // Once the innerModel built from the liteGraph successfully, the innerModel
     // owns the liteGraph, in which case, the invoker should not delete
diff --git a/frameworks/native/neural_network_runtime/nncompiler.cpp b/frameworks/native/neural_network_runtime/nncompiler.cpp
index 56ab88632363f1d65b9d2a3514243a5a4cd9f8a4..ad383f46d0e147cf486bd3f3d0b6156701945db3 100644
--- a/frameworks/native/neural_network_runtime/nncompiler.cpp
+++ b/frameworks/native/neural_network_runtime/nncompiler.cpp
@@ -130,6 +130,7 @@ NNCompiler::NNCompiler(const void* model, std::shared_ptr<Device> device, size_t
     m_modelName = m_innerModel->GetModelName();
     m_isProfiling = m_innerModel->GetProfiling();
     m_opLayouts = m_innerModel->GetOpLayouts();
+    m_tuningStrategy = m_innerModel->GetTuningStrategy();
 }
 
 NNCompiler::~NNCompiler()
@@ -375,7 +376,7 @@ OH_NN_ReturnCode NNCompiler::NormalBuild()
     }
 
     ModelConfig config {m_enableFp16, static_cast<OH_NN_PerformanceMode>(m_performance),
-        static_cast<OH_NN_Priority>(m_priority), m_isProfiling, m_cachePath, m_opLayouts};
+        static_cast<OH_NN_Priority>(m_priority), m_isProfiling, m_cachePath, m_opLayouts, m_tuningStrategy};
     if (m_liteGraph != nullptr) {
         ret = m_device->PrepareModel(m_liteGraph, m_quantBuffer, config, m_preparedModel);
     }
diff --git a/frameworks/native/neural_network_runtime/nncompiler.h b/frameworks/native/neural_network_runtime/nncompiler.h
index b5655a72f99ceb3b797f69d35ced10ae805b1b50..62b88ee672c80d449171b809306e6ebc6ea0db98 100644
--- a/frameworks/native/neural_network_runtime/nncompiler.h
+++ b/frameworks/native/neural_network_runtime/nncompiler.h
@@ -89,6 +89,7 @@ private:
     std::shared_ptr<mindspore::lite::LiteGraph> m_liteGraph {nullptr};
     std::vector<std::pair<std::shared_ptr<TensorDesc>, OH_NN_TensorType>> m_inputTensorDescs;
     std::vector<std::pair<std::shared_ptr<TensorDesc>, OH_NN_TensorType>> m_outputTensorDescs;
+    TuningStrategy m_tuningStrategy{TuningStrategy::OFF};
 };
 } // NeuralNetworkRuntime
 } // OHOS