Accelerating Inference - Towards a Full Language, Compiler and Hardware Stack