In search of effective and efficient Pipeline for Distillating Knowledge in Convolutional Neural Networks