Layer-Level Knowledge Distillation for Deep Neural Network Learning