Chronic Kidney Disease (CKD) is a progressive medical condition. The detection of disease in its early stages leads to severe complications, including kidney failure and mortality (sometimes it leads to patients dying). The study proposes a DL-based method for classifying CKD using real-time clinical ultrasound images to predict CKD. We use the DL Models like Dense-Net 121 (Densely Connected Convolutional Network) and VGG-Net to identify the subtle and hidden patterns in the medical images of the patients and classify whether the patient has CKD or not. This proposed model helps to predict the disease more accurately compared to other machine learning models and deep learning models add activation functions to add nonlinearity so that it can find more complex patterns. We collected ultrasound scans of CKD and non-CKD patients from various hospitals in Andhra Pradesh to employ cross-validation. Various data augmentation and preprocessing techniques, such as cropping, flipping, and resizing, help the models to learn more effectively. The results of this study achieve an accuracy of 97.67% using Dense-Net121 and 94.67% using VGG19 in classifying the CKD cases and non-CKD cases on testing data. Between the two applied models, Dense-Net121 performs best by achieving a high-test performance by addressing limitations like using a limited dataset, data imbalance, and model performance.