Bone fracture classification is a critical task in medical imaging, with automated approaches offering the potential to support timely and accurate diagnosis in clinical settings. In this study, we propose a lightweight convolutional neural network (CNN) model for automated bone fracture classification across ten types of fractures, including hairline, pathological, avulsion, comminuted, fracture dislocation, impacted, longitudinal, oblique, spiral, and greenstick fractures. Our model leverages depthwise separable and atrous convolutions, along with pooling layers, to capture multi-scale features effectively while maintaining a low computational footprint. With only 29,530 parameters, the model achieved an impressive accuracy of 98.04%, outperforming existing methods with significantly larger architectures. This efficiency makes the proposed model highly suitable for real-time deployment in resource-constrained clinical environments. Comparative analysis demonstrates that our model offers a superior balance between accuracy and efficiency, positioning it as a feasible solution for rapid fracture classification. Future research directions include enhancing model robustness to diverse imaging conditions and expanding its applicability to a broader range of fracture types.