Caffeine: Towards Uniformed Representation and Acceleration for Deep Convolutional Neural Networks