Data-intensive science applications in fields such as e.g., bioinformatics, health sciences, and material discovery are becoming increasingly dynamic and demanding with resource requirements. Researchers using these applications which are based on advanced scientific workflows frequently require a diverse set of resources that are often not available within private servers or a single Cloud Service Provider (CSP). For example, a user working with Precision Medicine applications would prefer only those CSPs who follow guidelines from HIPAA (Health Insurance Portability and Accountability Act) for implementing their data services and might want services from other CSPs for economic viability. With the generation of more and more data these workflows often require deployment and dynamic scaling of multi-cloud resources in an efficient and high-performance manner (e.g., quick setup, reduced computation time, and increased application throughput). At the same time, users seek to minimize the costs of configuring the related multi-cloud resources. While performance and cost are among the key factors to decide upon CSP resource selection, the scientific workflows often process proprietary/confidential data that introduces additional constraints of security postures. Thus, users have to make an informed decision on the selection of resources that are most suited for their applications while trading off between the key factors of resource selection which are performance, agility, cost, and security (PACS). Furthermore, even with the most efficient resource allocation across multi-cloud, the cost to solution might not be economical for all users which have led to the development of new paradigms of computing such as volunteer computing where users utilize volunteered cyber resources to meet their computing requirements. For economical and readily available resources, it is essential that such volunteered resources can integrate well with cloud resources for providing the most efficient computing infrastructure for users. In this dissertation, individual stages such as user requirement collection, user's resource preferences, resource brokering and task scheduling, in lifecycle of resource brokering for users are tackled. For collection of user requirements, a novel approach through an iterative design interface is proposed. In addition, fuzzy interference-based approach is proposed to capture users' biases and expertise for guiding their resource selection for their applications. The results showed improvement in performance i.e. time to execute in 98 percent of the studied applications. The data collected on user's requirements and preferences is later used by optimizer engine and machine learning algorithms for resource brokering. For resource brokering, a new integer linear programming based solution (OnTimeURB) is proposed which creates multi-cloud template solutions for resource allocation while also optimizing performance, agility, cost, and security. The solution was further improved by the addition of a machine learning model based on naive bayes classifier which captures the true QoS of cloud resources for guiding template solution creation. The proposed solution was able to improve the time to execute for as much as 96 percent of the largest applications. As discussed above, to fulfill necessity of economical computing resources, a new paradigm of computing viz-a-viz Volunteer Edge Computing (VEC) is proposed which reduces cost and improves performance and security by creating edge clusters comprising of volunteered computing resources close to users. The initial results have shown improved time of execution for application workflows against state-of-the-art solutions while utilizing only the most secure VEC resources. Consequently, we have utilized reinforcement learning based solutions to characterize volunteered resources for their availability and flexibility towards implementation of security policies. The characterization of volunteered resources facilitates efficient allocation of resources and scheduling of workflows tasks which improves performance and throughput of workflow executions. VEC architecture is further validated with state-of-the-art bioinformatics workflows and manufacturing workflows.