Deep neural networks have managed to outperform other machine learning algorithms. They have been proven to obtain great results in the field of image classification learning and recognition. Due to the high capability of representation learning by deep neural networks, many researchers in both academia and industry have attempted to apply it to many services in various areas. However, the large scale of real problems for abundant image databases causes high complexity and many computations for image learning and recognition. A new generation environment of distributed DNNs is needed so that the technology of managing DNNs is essential for making such intelligent information systems more effective. In this study, we first clarify the advantages and disadvantages of deep neural networks in terms of scalability, performance, computational power, and benefits of utilizing legacy DNNs in multiple DNN environments. Then, we propose a method for indexing and finding deep neural networks for image recognition. Our method is a new architecture for selecting one DNN from a simple DNN to answer the query image automatically. Our method consists of three essential features: (1) a specific DNN architecture for specific domain recognition, (2) a training model for specific DNN and aggregation for global meta-DNN, and (3) image classification for an image query to global meta-DNN. In several experiments using multiple DNNs with different domains of image recognition, we evaluate the feasibility of our proposed method.