A highly dangerous cancer, which needs invasive, costlier and time-consuming diagnostic tests are considered in this study. ALL diagnosis based on peripheral blood smear (PBS) images are taken for the implementation. There exist chances of misclassifying the cancer cases by the medical experts. Therefore, an appropriate diagnosing model was developed based on Deep learning algorithms to distinguish ALL subtypes. CNNs are the most prominent model to use in classifying bio-medical images with high percentage of accuracy. For this study, an ensemble model was built by having a publicly available ALL data set, which consists of 3256 PBS images. Three CNN architectures were used to classify the 4 class labels. The proposed model has been evaluated using a 5-fold cross-validation scheme. It achieves a classification accuracy of 99.23%.