Yoga pose classification in videos plays a crucial role in fitness and wellness applications, requiring efficient and adaptable models for accurate recognition. This paper introduces a meta-learning approach using Prototypical Networks for classifying yoga poses in video datasets. The methodology involves feature extraction through deep neural networks (DNNs) which are fed into Prototypical Network for computation of prototypes. These prototypes represent each yoga pose in the feature space and are used for classification based on similarity scores computed using distance metrics. The model undergoes episodic training, optimizing parameters for quick adaptation to new tasks with limited data. Experimental results on a dataset comprising 88 videos with 5 yoga poses demonstrate the effectiveness of the proposed approach. In a 5-way 20-shot setting using Prototypical Networks, the authors achieved GFLOPs of 17.75418, 6.95% gain in accuracy, 85% reduction in inference time, and 30.56% reduction in the model size over the traditional DNN-based methodology. The paper concludes by highlighting the potential of meta-learning techniques in enhancing fitness and wellness applications.