ReffAKD: Resource-efficient Autoencoder-based Knowledge Distillation