Accurate bone fracture detection can be critical in the medical diagnosis process. This paper proposes a bone fracture detection approach based on deep learning using Convolutional Neural Networks (CNNs) with X-ray images. We have explored three models: one predefined model (MNASNet1) and two custom models (Proposed Model 1 and Proposed Model 2). Model performance is evaluated using metrics such as F1-Score, precision, and accuracy. Results show that the proposed models achieved outstanding performance compared to the MNASNet1 model in most of the metrics. Specifically, Proposed Model 1 achieves the highest overall F1-score, recall, and precision (1.00) and also has the highest validation accuracy (99.89%). Our objective is to develop a custom CNN architecture that demonstrates the potential of our proposed models to enhance fracture detection accuracy and develop a web application to detect bone fractures faster to facilitate more precise clinical assessment and treatment decisions. Doctors can emphasize the benefits and explore collaboration opportunities in a relatable way to integrate our web application into medical practices.