diff --git a/1.1/HalPolicy.cpp b/1.1/HalPolicy.cpp
index dbd380a..9f2c9be 100644
--- a/1.1/HalPolicy.cpp
+++ b/1.1/HalPolicy.cpp
@@ -297,10 +297,11 @@
         return Fail("%s: Could not read output", __func__);
     }
 
-    const armnn::TensorInfo& outputInfo = GetTensorInfoForOperand(*output);
+    armnn::TensorInfo outputInfo = GetTensorInfoForOperand(*output);
     if (IsDynamicOutput(outputInfo))
     {
-        return Fail("%s: Dynamic output not supported", __func__);
+        ALOGD("Output shape not set, will infer from inputs");
+        outputInfo.SetShape(InferPadOutputShape(inputInfo.GetShape(), descriptor.m_PadList));
     }
 
     if (!IsLayerSupportedForAnyBackend(__func__,
@@ -318,7 +319,12 @@
     input.Connect(layer->GetInputSlot(0));
     layer->GetOutputSlot(0).SetTensorInfo(outputInfo);
 
-    return SetupAndTrackLayerOutputSlot<hal_1_1::HalPolicy>(operation, 0, *layer, model, data);
+    return SetupAndTrackLayerOutputSlot<hal_1_1::HalPolicy>(operation,
+                                                            0,
+                                                            *layer,
+                                                            model,
+                                                            data,
+                                                            armnn::Optional<armnn::TensorInfo>(outputInfo));
 }
 
 bool HalPolicy::ConvertSpaceToBatchNd(const Operation& operation, const Model& model, ConversionData& data)
diff --git a/1.2/HalPolicy.cpp b/1.2/HalPolicy.cpp
index 58fcf73..cdf8c0f 100644
--- a/1.2/HalPolicy.cpp
+++ b/1.2/HalPolicy.cpp
@@ -453,12 +453,6 @@
         return Fail("%s: Could not read output", __func__);
     }
 
-    const armnn::TensorInfo& outputInfo = GetTensorInfoForOperand(*output);
-    if (IsDynamicOutput(outputInfo))
-    {
-        return Fail("%s: Dynamic output not supported", __func__);
-    }
-
     const armnn::TensorInfo& inputInfo  = input.GetTensorInfo();
     unsigned int rank = inputInfo.GetNumDimensions();
 
@@ -468,6 +462,13 @@
         return Fail("%s: Could not convert paddings", __func__);
     }
 
+    armnn::TensorInfo outputInfo = GetTensorInfoForOperand(*output);
+    if (IsDynamicOutput(outputInfo))
+    {
+        ALOGD("Output shape not set, will infer from inputs");
+        outputInfo.SetShape(InferPadOutputShape(inputInfo.GetShape(), descriptor.m_PadList));
+    }
+
     // Determine type of padding value
     OperandType operandType0;
     OperandType operandType2;
@@ -528,7 +529,12 @@
     input.Connect(layer->GetInputSlot(0));
     layer->GetOutputSlot(0).SetTensorInfo(outputInfo);
 
-    return SetupAndTrackLayerOutputSlot<hal_1_2::HalPolicy>(operation, 0, *layer, model, data);
+    return SetupAndTrackLayerOutputSlot<hal_1_2::HalPolicy>(operation,
+                                                            0,
+                                                            *layer,
+                                                            model,
+                                                            data,
+                                                            armnn::Optional<armnn::TensorInfo>(outputInfo));
 }
 
 bool HalPolicy::ConvertPrelu(const Operation& operation, const Model& model, ConversionData& data)
diff --git a/OutputShapeUtils.cpp b/OutputShapeUtils.cpp
index a0c624c..a1a49d7 100644
--- a/OutputShapeUtils.cpp
+++ b/OutputShapeUtils.cpp
@@ -6,6 +6,7 @@
 #include "OutputShapeUtils.hpp"
 
 #include <algorithm>
+#include <vector>
 
 namespace armnn_driver
 {
@@ -17,6 +18,24 @@
     return outputInfo.GetNumElements() == 0u;
 }
 
+TensorShape InferPadOutputShape(const TensorShape& inputShape,
+                                const std::vector<std::pair<unsigned int, unsigned int>>& padList)
+{
+    const unsigned int numDims = inputShape.GetNumDimensions();
+
+    std::vector<unsigned int> outputDims;
+    TensorShape outputShape = TensorShape(numDims);
+    for (unsigned int dim = 0; dim < numDims; ++dim)
+    {
+        unsigned int dimSize = inputShape[dim];
+        const std::pair<unsigned int, unsigned int>& dimPadding = padList[dim];
+        dimSize += dimPadding.first;
+        dimSize += dimPadding.second;
+        outputShape[dim] = dimSize;
+    }
+    return outputShape;
+}
+
 TensorShape InferPreluOutputShape(const TensorShape& inputShape, const TensorShape& alphaShape)
 {
     // NOTE: The inferred PReLU output size will be the maximum size along each dimension
diff --git a/OutputShapeUtils.hpp b/OutputShapeUtils.hpp
index 6e2a76d..d1818db 100644
--- a/OutputShapeUtils.hpp
+++ b/OutputShapeUtils.hpp
@@ -12,6 +12,9 @@
 
 bool IsDynamicOutput(const armnn::TensorInfo& outputInfo);
 
+armnn::TensorShape InferPadOutputShape(const armnn::TensorShape& inputShape,
+                                       const std::vector<std::pair<unsigned int, unsigned int>>& padList);
+
 armnn::TensorShape InferPreluOutputShape(const armnn::TensorShape& inputShape, const armnn::TensorShape& alphaShape);
 
 } // namespace armnn_driver