Automatic product recognition via product image scan has a positive impact for both economic and social progress is it faster than human based product identification and more reliable. Object recognition via images has become popular in the field of computer vision due to the great application prospect, such as automatic product checkout, goods management, and stock tracking. As retail is evolving, companies are increasingly focusing on the integration of AI technology in the day-to-day activities of a retail store. The research study presented in this paper aims to investigate the use of deep learning models such as Convolutional Neural Network (CNN), ResNet50, and VGG16 to classify a large set of toy images representing items from a toys retail store. The performance of the three models investigated was analyzed in terms of training accuracy, training validation, training loss, average runtime per epoch and test accuracy. The dataset contains over 21,000 toy images. The study consisted of two testing scenarios that used a 70:30 data split ratio and 80:20 ratio respectively, for training and testing. Both VGG16 and ResNet50 modules provided very similar accuracy for both scenarios and outperformed the CNN model. However, a 9.3% and 14.2% increase in the average runtime per epoch was observed in the 80:20 scenario for the VGG16 model and ResNet50 model respectively. Hence, it was concluded that the best runtime per epoch and accuracy was achieved when the data is split into 70:30 train test ratio and ResNet50 model produced the best results.