Facial Emotion Recognition (FER) and Human-Centered Software Engineering (HCSE) relate to each other whereby systems designed using HCSE will be able to adapt themselves in response to the emotional states of a user. The paper proposes the deep learning approach to emotion recognition according to the customer service application. The approach makes emotion classification simpler by re-classifying the basic emotions into three behavioral categories namely satisfied, neutral and dissatisfied. Two models are used, a custom made Convolutional Neural Network (CNN) and a VGG16 model fine-tuned by transfer learning. The two were trained and tested on a variant of the CK + dataset. The preprocessing flow is as follow: grayscale conversion to normalization and data augmentation, which helps to create a more robust and more general model. It is observed in the experimental results that the custom CNN performs better than VGG16 and a test accuracy of 97% was received with better performance in terms of precision, recall, and F1-score. The data contained in confusion matrix proves the good performance of the model in detecting fine emotional cues that are applicable in the service settings. The experiment illustrates the importance of effective, domain-specific models and streamlined representation of emotion during real time human-computer interaction. Privacy and bias are ethical issues that are recognized. Future efforts will be conducted on multimodal emotion recognition and better model explainability and real-world applications in the customer service setting.