Cancer takes many lives in worldwide, in 2022, colorectal cancer is the 2nd cause of the greatest number of deaths (903,859 deaths). Pathologists find it difficult to accurately diagnose lesions based on biopsy characteristics under naked eye microscope, thus the diagnosis differs from one pathologist to another. Furthermore, the availability of highly qualified personnel is limited, increasing patient waiting times and potentially delaying life-saving treatments. For this, it is necessary to develop automated methods to improve the accuracy of diagnosis, reduce delays, and improve patient outcomes. This study explores and compares advanced deep learning models (ResNet50, VGG16, DenseNet121, EfficientNetB0, InceptionV3), a custom CNN, a Random Forest classifier, and a hybrid CNN-Random Forest model to classify histopathological images of colorectal cancer (CRC). Experimental results show that ResNet50 offers the highest accuracy (95%), highlighting the potential of deep learning models, to support pathologists with more reliable and consistent diagnostic tools.