The basic CAN spec does not define Channels. Essentially, a CAN network is a facility for passing messages around a group of devices. A message can be transmitted by any device, and is available to every device. A message has two parts: its Identifier and some Data. That's CAN in a nutshell.
There are higher level protocols built on top of CAN (e.g. CANopen) that give meanings to messages: defining what kind of data they contain, and how and when these message get on the network. These protocols typically add features, such as node addresses, network masters, and block data transfers, that are not part of the basic CAN spec.
"Channel" is a somewhat equivocal term; its meaning depends on the context. Broadly, a channel is some kind of conduit for data. For example, incoming messages might be directed into channels that are used by different processes. This way a process wouldn't need to search through all received messages to look for ones that are relevant, it would only need to check for messages in its channel.
I'm not familiar with Microchip Harmony, but from the description of the CAN interface it looks like "channel" essentially means "buffer". It appears that received messages are sorted into channels (message buffers) based on which acceptance filter it passes through. So you can direct messages to different buffers based on the message identifier. Channels are used for transmit as well, as a mechanism for buffering and setting relative priority of outgoing messages. The channel is not part of the actual message, so it doesn't have any meaning at the network level.