This study conducts an empirical investigation focused on enhancing the performance of automatic modulation classification in wireless communication systems. We integrate a Customized CNN model with deep learning features. Initially, we employ pre-deep learning models to extract crucial features. These extracted features are then fed into a custom CNN classifier for precise modulation format categorization. To provide a comprehensive comparison of feature classification against pre-deep learning models, we find that our proposed model achieves superior classification results. We validate our approach using the publicly available RADIOML 2018.01A dataset. Our experiments reveal that our proposed scheme achieves an impressive accuracy of 95.16%, outperforming other state-of-the-art classification methods.