The subtyping of renal cell carcinoma (RCC) is essential for accurate diagnosis and therapy planning. In order to improve the automated histological subtyping of RCC, this work presents a hybrid deep learning architecture that combines ResNet50 and Vision Transformers (ViT). Using the advantages of both architectures, our method combines global contextual knowledge from ViT with local feature extraction from ResNet50. With an average cross-validation accuracy of 93.47% and a test accuracy of 93.38%, experimental results show a notable increase over baseline models. The model uses stratified cross-validation and transfer learning to overcome the difficulties caused by high intra-class variability and a lack of medical datasets. The suggested methodology bridges the gap between clinical application and computational breakthroughs by helping pathologists with precise RCC subtyping while demonstrating promising diagnostic performance.