Codon optimization is the process of redesigning mRNA sequences to improve gene expression efficiency in a given host organism. It plays a crucial role in recombinant protein production and therapeutic mRNA design, where both translation efficiency and structural stability determine the final level of protein expression. However, conventional optimization approaches often rely on predefined codon usage tables or heuristic rules, making it difficult to achieve a consistent balance between expression efficiency and mRNA stability across diverse organisms. To address this limitation, this study presents a deep learning framework for codon optimization that aims to enhance both protein expression and mRNA stability. The framework, combining Transformer and 1D CNN architectures with a joint loss function to balance these two biological x-objectives, formulates codon optimization as a sequence optimization task that learns to generate biologically efficient mRNA sequences from amino acid inputs. This balanced optimization enables rational and data-driven gene design, providing a robust computational approach for improving translational performance and structural robustness. Ultimately, the proposed framework has the potential to maximize protein expression while maintaining mRNA stability, offering a scalable foundation for advanced therapeutic mRNA and synthetic biology applications.