Not KVM bound. The VFIO API deconstructs a device into regions, irqs, etc. The userspace application (QEMU, cloud-hypervisor, etc..) is responsible for reconstructing it into a device for e.g. a guest VM to consume.
Boot with intel_iommu=on.
Devices are bound together for isolation, IOMMU capabilities and platform topology reasons. It is not configurable.