Colorectal cancer is a serious and common disease that needs accurate and timely diagnosis to produce better treatment outcomes. Deep learning and computer vision techniques are employed in this study to enhance the detection and classification of colorectal cancer. Various convolutional neural network (CNN) models-like well-known architectures like VGG16, VGG19, AlexNet, NASNet, and ResNet-50-were utilized and tuned, and a CNN specifically developed for this purpose. Among all the models compared, custom CNN exhibited the highest performance with a training accuracy of 97.47 %, test accuracy of 98.93 %, sensitivity of 99.75 %, and a perfect specificity of 100 %. Models like VGG16 and VGG19 also provided high performances with a test accuracy of 95.42 % and 94.78 % respectively, both having a sensitivity of 100 %, but their specificities were slightly lower at 99.75 % and 98.73 %. ResNet-50, however, showed a sharp drop in performance, wherein test accuracy went down to 57.93 % and specificity came down to 62.66 %, while training accuracy was high at 93.14 %, which is an indication of overfitting. These results confirm the effectiveness of the introduced custom CNN in correctly identifying colorectal cancer lesions, thus offering clinicians an accurate and automatic diagnostic tool. The incorporation of deep learning into medical imaging workflows can greatly accelerate early detection and patient outcomes. Scaling up the size and diversity of annotation datasets and exploring alternative deep learning architectures should be the subject of future work, aiming to increase diagnostic accuracy further.