Nowadays, many parents struggle to comprehend their children's emotions, which can hinder the creation of a nurturing environment. While numerous models focus on predicting adult emotions, there is a lack of standardised datasets for studying children's emotions. To address this gap, our work attempts to establish a comprehensive children's emotion dataset that can facilitate the study of emotions across various pose orientations. Furthermore, we propose an efficient and deployable system for real-time children's emotion prediction. An effective face detector with deep architecture is designed to handle all pose orientations from key image frames. Optimal features are then selected by re-ranking the features using a hybrid feature selection mechanism. The emotion category is declared by carefully analysing sequences of emotion identification from these features. This system holds promise for educational institutions and healthcare facilities, offering insights into children's behaviour through emotional analysis. Through experimental comparisons with three state-of-the-art emotion prediction models, we observed that our proposed system consistently outperforms existing models. Hence, we strongly recommend the adoption of our proposed system. With its achievement of state-of-the-art results in children's facial emotion recognition, it offers a practical solution for real-time deployment across diverse settings.