If I'm correct you'll need to replace all HTML tags with their corresponding 'marks/signs' (& for example becomes &).
You could achieve this with a custom java action (Unescape HTML - more or less like the EscapeHTML in the CommunityCommons), you can find an example below:
// BEGIN USER CODE return HTML.replace(""", "\"") .replace("&", "&") .replace("<", "<") .replace(">", ">") .replace("'", "'") .replace(" ", " ") .replace("+", "+"); // END USER CODE
You can also try to resolve this with normal microflow actions using the replace or replaceAll functions, but I think the custom java action will get you there more easily.
Regarding the images: this is something I haven't tried myself yet, so maybe someone else could give some more advice on this.
There are some standard modules to convert text / images from HTML to plain / filedocuments.
You could try and use the "HTMLToPlainText" java action from the CommunityCommons module.
Images are either uploaded as a document or as a base64 string, you can convert the base64 string to an image with the "Base64DecodeToFile" Java action, also in the CommunityCommons module.
Directly from CKEditor to Word isn't a Mendix standard, you could try and use the CommunityCommons Module in combination with the DocumentTemplates or write a Java action to create this for you, based on the html string from the CKEditor.