Tomato crop health is frequently compromised by various leaf diseases, leading to significant yield losses if not detected in time. This paper presents a conceptual design of a hybrid deep learning architecture aimed at enhancing disease classification accuracy by integrating Convolutional Neural Networks (CNN) and Vision Transformers (ViT). The model utilizes EfficientNet for local feature extraction and ViT to capture broader spatial dependencies within tomato leaf images. An attention-based feature fusion mechanism is proposed to dynamically prioritize regions affected by disease. Data augmentation and focal loss are incorporated to improve robustness and address class imbalance. While empirical results are not yet available, this work lays the foundation for a practical and efficient disease detection framework in precision agriculture, with future validation and deployment planned.