This study introduces a novel deep learning framework, the Deep Convolutional K-Means (DCKM), designed to cluster 3D morphologies of human legs for applications in medical compression stockings and adaptive leg morphotypes. By leveraging Convolutional Neural Networks (CNNs) and integrating K-means clustering into the Deep Convolutional Transform Learning (DCTL) framework, the method eliminates the need for a decoder network, reducing overfitting risks and computational complexity. The proposed method efficiently processes 3D point clouds generated from mesh representations of human legs, identifying four distinct morphological clusters. Each cluster is described based on geometric and anthropometric landmarks, enabling more accurate classification of leg shapes. The results highlight the potential of DCKM in medical applications, especially for improving compression stocking designs by adapting to specific leg morphologies. Despite its success, challenges related to transparency and interpretability in unsupervised clustering remain. Future research will focus on enhancing the interpretability of DCKM and expanding the dataset to include pathological leg shapes.