Cancer diagnosis in different organs of the human body is an area of interest for many scientists around the world. Gallbladder cancer is a disease that needs to be diagnosed at the early stages of its occurrence. Computer Aided Diagnosis (CAD) based systems provide support to healthcare practitioners in early diagnosis by providing second opinion. Deep convolutional neural networks (DCNNs) are the backbone of CAD system and provide strong capabilities in distinguishing types of gallbladder disease from ultrasound images. In this work, a customized DCNN architecture named Residual-Dense 126 (DepthNet126) is proposed for classification of gallbladder disease from gallbladder ultrasound images. The proposed architecture is based on the residual and dense mechanism, where the layers are added along with skip connection that later combined using a depth concatenation layer. A tailored dataset named Gallbladder Cancer Ultrasound (GBCU) is employed to conduct experiments. In the training phase, hyperparameters are initialized through Bayesian Optimization. The trained model is finally tested on the test set images and demonstrates an accuracy of 87.80% in classifying the disease in underlying types. The results are compared with state-of-the-art (SOTA) techniques and show the proposed architecture significant improvement in classifying the intrinsic features of medical images.