Biometrics representing biological and physiological characteristics of an individual which are used to authenticate individuals are becoming increasingly common in many areas such as banking, airport transfer, document authorization, cyber forensics, land registration etc. Among many biometric traits, iris biometric is considered the most reliable and secure metric due to its universality, uniqueness, permanence, and collectability. It is noted by the researchers that the iris texture also varies between the left and right eye and even between twins, making it a highly secure authentication method. The existing iris recognition methods address many challenges such as occlusion, scaling, rotation, motion blur, pupillary dilation, irregular reflections in non-cooperative environments, etc. In order to improve the accuracy or robustness against these challenges, deep-learning-based person authentication methods have been developed that accurately identify genuine and imposter by analyzing the differences between corresponding patches in pairs of iris images. The convolution neural network (CNN) based VGG 16, Inception, Nasnet, and Mobilenet deep learning models provide the best recognition performance on different iris datasets. Motivated by the fact that the ensemble algorithms significantly improve the accuracy of any computer vision systems, in our work, we have designed an ensemble model that combines the best features of deep learning models. The ensemble model uses techniques such as stack ensemble, voting, and bagging with classifiers such as Logistic regression, random forest, decision tree, and K nearest neighbour to improve the classification performance of the iris recognition system. Our experiments using four well-known datasets (CASIA, Polaris IITD, Ubiris, and MMU) have demonstrated the accuracy of the proposed ensemble model designed with CNN-based deep learning algorithms.