The posture estimate technique has been widely applied in physical therapy and sports fields, but this technique is scarcely used in the yoga domain. Classifying yoga positions using neural networks is a challenge due to the inherent complexity of yoga poses compared to regular stances. This paper presents a yoga pose classification where a skeleton of posture is obtained using body joint location. Then, the obtained skeleton is imposed on the original yoga pose image. Later, the transfer learning approach of convolutional neural network using VGG19 is trained and used using skeleton-imposed images for posture classification. Model accuracy, loss, and the confusion matrix of individual yoga poses were used to evaluate the classification results. Furthermore, the confusion matrix is diversified in terms of precision, recall, and F1-score. With the most joints in the body, the proposed model succeeded with a classification accuracy of 96%.