The main cause of cancer-related death worldwide is still lung cancer. Globally, lung cancer is the primary cause of cancer-related death, and it is quite difficult to make an early and precise diagnosis of the lung cancer. Non-small cell lung cancer (NSCLC) is one of the subtypes of lung cancer, that requires novel methods for accurate detection and classification. When examining big collections of histopathological images, the accuracy and scalability of current NSCLC identification techniques are limited. Current advancements in deep learning have shown great promise for the processing of histopathological images. The three categories of NSCLC, squamous cell carcinoma, adenocarcinoma and large cell carcinoma are the main focus of this study. The LC25000 dataset, which includes 25,000 high-resolution images, is used in this work to develop an ensembled model that combines four models named custom CNN, ResNet-50 V2, and Inception ResNet V2. To enhance the quality of the data, a lot of preprocessing was used, including augmentation and normalization. The model was evaluated on the basis of precision, F1-score, recall, and accuracy. The ensemble methods outperform individual architectures in classification challenges with the 99.89% accuracy. Future studies can concentrate on using self-supervised learning for data-efficient training, incorporating domain adaption strategies to enhance generalization across various datasets.