The study evaluates how well Convolutional Neural Networks (CNNs) identify dog breeds with deep learning technology. Majority of the families across the global population considers their pets as the integral part of the family. It’s therefore the duty of the various pet owners to ensure that they take good care of the animals. To be precise, out of all pets’; dogs are those that are most adored. Reading emotions of dogs is not easy for man and this involves understanding of the abnormal physical and mental state of the dog. Recently, Convolutional Neural Networks (CNNs), have proved themselves as a powerful tool for solving these challenges by allowing them to automatically learn discriminative features from images. More than 10,000 images spanning 120 different breeds of dogs served as the dataset to evaluate ResNet50 and InceptionV3 and a user-built CNN architecture for training and testing. The most efficient model turned out to be InceptionV3 which reached a 98.80% accuracy level better than both the custom CNN (83.33%) and ResNet50 (58.89%). This study evaluates the relationship between model performance outcomes caused by transfer learning and network depth together with data augmentation practices. The research delivers three main contributions which consist of (1) evaluating different CNN models for breed classification tasks along with (2) implementing transfer learning for accuracy enhancement and also (3) developing a systematic approach to data preprocessing and augmentation to boost model generalization capabilities.