Parameter-Efficient Tuning of Large Language Models (LLMs) with Novel Ensemble Knowledge Distillation Framework for NLP