In recent years, graph neural networks (GNNs) have been widely adopted in representation learning of graph-structured data and provided state-of-the-art performance in various application such as link prediction and node classification. Simultaneously, self-supervised learning has been studied to some extent to leverage rich unlabeled data in representation learning on graphs. However, employing self-supervision tasks as auxiliary tasks to assist a primary task has been less explored in the literature on graphs. In this paper, we propose a novel self-supervised auxiliary learning framework to effectively learn graph neural networks. Moreover, we design first a meta-path prediction as a self-supervised auxiliary task for heterogeneous graphs. Our method is learning to learn a primary task with various auxiliary tasks to improve generalization performance. The proposed method identifies an effective combination of auxiliary tasks and automatically balances them to improve the primary task. Our methods can be applied to any graph neural networks in a plug-in manner without manual labeling or additional data. Also, it can be extended to any other auxiliary tasks. Our experiments demonstrate that the proposed method consistently improves the performance of link prediction and node classification on heterogeneous graphs.