Potholes pose a persistent threat to road safety, infrastructure durability, and commuter wellbeing. This situation is worse in developing countries where maintenance resources are limited. To address this challenge, this study developed a custom lightweight Convolutional Neural Network (CNN) architecture to classify road surface images as either pothole or normal. Our model was trained and evaluated on a dataset consisting of 5000 balanced images of normal road conditions and potholes. Data augmentation was applied to enhance the diversity of the training images, and early stopping was used a technique to improve generalizability, avoiding overfitting. The proposed model achieved a validation accuracy of 0.91 and an F1-score of 0.90, demonstrating robust performance on unseen images captured in Mauritius. Precision for pothole detection reached 0.96, indicating strong classification ability with minimal false positives. These results underscore the effectiveness of our CNN based approach in supporting automated road condition monitoring, ready for deployment onto mobile devices. This study, as research in progress, covers the development of the lightweight model, while the deployment is still under consideration.