Neural networks have given rise to new types of models for solving diverse challenges in the field of Artificial Intelligence. However, these networks suffer from issues such as reliance on sequential data, vanishing or exploding gradients, and computational intensity, to mention a few. Many neural networks have been proposed in the past to bypass these issues and among them Random Vector Functional Link Neural Networks (RVFLNNs) have shown distinct performance. RVFLNNs belong to the category of feed-forward neural networks and have been prevalently used in many areas of research due to their ability to act as universal approximators. RVFLNN’s main characteristics include fast training speed, fairly simpler structure, direct links, and better optimization. RVFLNNs have been developed with various base architectures, such as shallow networks, ensemble/deep networks, distributed networks, semi-supervised /unsupervised networks, and forecasting networks, each serving a unique purpose. They are widely being used to solve different problems pertaining to classification, regression and prediction tasks. This paper aims to give the readers a good understanding of these networks and also provide a summary of various cutting-edge RVFLNN models in terms of their architectural differences and applications. This survey has shed light on existing problems with RVFLNNs such as its inability to work with imbalanced data, lack of proper channels to support sequential processing, and lack of suitable distributed learning models. Lastly, potential future research opportunities to convalesce the architecture and learning algorithms in the development of efficient RVFLNNs are discussed.