The Kidney disease has been recognized as a worldwide public health problem. This study represents the powerful and unique Kcst_Net architecture to enhance the automatic detection and classification of kidney disease in CT images. The proposed architecture included feature detectors that comprise a convolutional layer, max pooling layer with batch normalization. Dropout of 30% is included to avoid the overfitting problem. The features generated are flattened and are subjected to a dense layer classifying four classes: kidney cyst, kidney stone, kidney tumor and normal kidney. The model was analyzed with the data collected from Kaggle which has 9000 unique CT images for four class classifications. The dataset was trained and tested using RMSProp optimizer, which outperformed with 99.86% testing accuracy. The experimental results show that the proposed Kcst_Net model attained the best accuracy in classification with 0.99, 0.98, and 0.98 as precision, recall, and F1-score. The proposed Kcst_Net model, with its ability to accurately categorize kidney disease categories from CT scans, can be an essential support to medical professionals in making clinical decisions for kidney disease management.