Deep learning models, renowned for their exceptional performance in various tasks, demand significant computational resources, making optimization techniques crucial for enhancing their efficiency. Pruning, quantization, and knowledge distillation are key methods for achieving this, each addressing specific challenges. Pruning reduces model size and complexity by eliminating less important neurons, potentially improving inference speed and lowering energy consumption. Quantization decreases memory usage and computation time by representing weights with lower numeric precision, suitable for deployment on a range of hardware, albeit with possible performance trade-offs. Knowledge distillation compresses models by transferring knowledge from a larger "teacher" model to a smaller "student" model, maintaining accuracy while supporting versatile designs. The choice of optimization technique depends on the model type, deployment environment, and performance goals, and these methods collectively aim to lessen the environmental impact of deep learning models.