Federated Learning (FL) presents a promising solution to the privacy concerns associated with deep learning-based recommender systems. Our work introduces FedxDeepFM, an advanced recommendation model that integrates deep learning with FL to enhance user privacy. FedxDeepFM operates on the principles of FL, conducting decentralized training on user devices and sharing only model parameters, thereby ensuring that no user data is exposed to the server. The model also incorporates privacy features, such as pseudo-interaction padding, which enhance its resistance to inference attacks. Evaluated on benchmark datasets, FedxDeepFM demonstrates exceptional recommendation quality while maintaining stringent privacy standards, outperforming contemporary models and addressing the common challenges such as sparsity and cold starts in recommendations.