https://docs.nvidia.com/deeplearning/sdk/tensorrt-developer-guide/index.html#ipluginext

In the doc, it is written below:

1. Create the Plugin and AddPlugin int the network.

For example, to add a plugin layer to your network with plugin name set to pluginName and version set to pluginVersion, you can issue the following:
//Use the extern function getPluginRegistry to access the global TensorRT Plugin Registry
auto creator = getPluginRegistry()->getPluginCreator(pluginName, pluginVersion);
const PluginFieldCollection* pluginFC = creator->getFieldNames();
//populate the field parameters (say layerFields) for the plugin layer
PluginFieldCollection *pluginData = parseAndFillFields(pluginFC, layerFields);
//create the plugin object using the layerName and the plugin meta data
IPluginV2 *pluginObj = creator->createPlugin(layerName, pluginData);
//add the plugin to the TensorRT network using the network API
auto layer = network.addPluginV2(&inputs[0], int(inputs.size()), pluginObj);
… (build rest of the network and serialize engine)
pluginObj->destroy() // Destroy the plugin object
… (destroy network, engine, builder)
… (free allocated pluginData)

2. For deserialization, the doc say:

During serialization, the TensorRT engine will internally store the plugin type, plugin version, and namespace (if it exists) for all IPluginV2 type plugins. During deserialization, this information is looked up by the TensorRT engine to find the Plugin Creator from the Plugin Registry. This enables the TensorRT engine to internally call the IPluginCreator::deserializePlugin() method. The plugin object created during deserialization will be destroyed internally by the TensorRT engine by calling IPluginV2::destroy() method.

In previous versions of TensorRT, you had to implement the nvinfer1::IPluginFactory class to call the createPlugin method during deserialization. This is no longer necessary for plugins registered with TensorRT and added using addPluginV2.