In this paper, we present HYPC-Net, a hybrid model that uses deep convolutional neural networks in combination with classical machine learning techniques to yield the highest performance only with RGB yoga pose images. HYPC-Net utilizes both convolutional features and classical classifiers that significantly increase the accuracy and robustness in yoga pose recognition. We conducted a variety of experiments with the Yoga-82 dataset to confirm that HYPC-Net is indeed effective. We present a comprehensive comparative analysis considering the existing literature and state-of-the-art CNN models. Specifically, the CatBoost version of HYPC-Net showed the highest accuracy and weighted F1 score of \(96.7\%\) for class-6. Moreover, the Random Forest variant of HYPC-Net had the best accuracy and weighted F1 scores of \(95.4\%\) and \(93.6\%\) for class-20 and class-82, respectively. Our results clearly indicate that combining deep learning with traditional machine learning techniques in a hybrid approach has paved a new way to explore yoga pose classification.