Class-Aware Pruning for Efficient Neural Networks